Communication Science Corporate Communication University of Twente
Supervisors: First: Dr J. J. van Hoof Second: Dr J. F. Gosselt Master thesis by
BSc S. van ‘t Slot
September 18th, 2017
Creating interaction in the online political sphere:
Which Twitter content triggers response?
2
3
Abstract
Introduction The popularity of social media platforms amongst citizens has led to the presence of politicians who use these platforms to generate their own content, and interact with their voters.
Especially interaction on social media is beneficial for politicians as it can be used to keep citizens up to date, to give citizens attention, to entice new people in politics, and to help increase youngsters’
political efficacy. However, despite the benefits, interaction remains unexplored in research and it remains unknown what kind of messages and message characteristics can trigger interaction.
Objective This study aimed at finding out which content from list pullers triggers interaction on Twitter. Method A content analysis of all tweets started by thirteen list pullers during the election period of the 2017 Dutch elections (N = 2158) was executed. The analysis consisted of three content categories: topic and issue (topics, visual content, and tweet characteristics), opinion and sentiment (tone, and humour), and structural category (actors). Results Results from the topic and issue category showed that more interaction is found in tweets with an international topic (e.g.
war/terrorism, or Europe) or a comment on the government and other politicians, whereas tweets with a national topic (e.g. education, or health) triggered less interaction. Tweets with visual content (carrying the corporate social identity of the politician’s party) triggered more interaction. The tweet characteristics hashtags and emoticons triggered more interaction, whereas @-mentions and URL’s led to less interaction. The opinion and sentiment category showed that a slightly negative tone triggers more interaction, and that humour appeals to Twitter users in the form of likes. The results found for the structural category only show that mentioning international politicians leads to more interaction. Conclusion If list pullers aim at triggering interaction with their tweets, they can gain most benefit from focusing on the topic and issue category. Implications for political communication will especially show in the social media strategies of politicians and the involvement of citizens in political communication. Political communication with a focus on interaction will change one-way broadcasting to a dynamic and complex conversation in which more people will be actively involved than ever before.
Keywords Political communication, social media, Twitter, interaction, content analysis
4
5
Table of Content
1. Introduction ... 7
2. Theoretical framework ... 8
2.1 Political communication ... 8
2.2 Political communication and social media ... 9
2.3 Interactivity ... 11
2.4 Aspects of interaction on Twitter ... 12
2.5 Research goal ... 17
3. Method ... 18
3.1 Context ... 18
3.2 Corpus ... 18
3.3 Codebook ... 19
3.4 Validity and reliability ... 21
3.5 Data analysis ... 22
4. Results ... 24
4.1 Topic and issue category ... 24
... 28
4.2 Opinion and sentiment category ... 29
4.3 Structural category ... 30
4.4 Additional content ... 32
5. Discussion ... 37
5.1 Interpretation of findings ... 37
5.2 Implications for political communication ... 38
5.3 Limitations ... 39
5.4 Conclusion ... 39
Literature ... 40
Appendices ... 45
6
7
1. Introduction
Social media platforms, such as Twitter, provide politicians with the possibility to reach their potential voters independently and personally in a direct manner, and on a regular basis (Vergeer & Hermans, 2013). As a consequence, many politicians are present on social media platforms (Stieglitz & Dang- Xuan, 2013). In practice, politicians mostly use these platforms for one-way communication (Vergeer
& Hermans, 2013).
However, using social media platforms as a tool to create interaction with or between citizens could make a substantial difference in elections. After all, Spierings and Jacobs (2014) suggest that interactivity might be the key to the hearts of voters. This is not only because interaction with politicians and other citizens fulfils citizens’ desire to be kept up to date and to receive attention (Spierings & Jacobs, 2014). Namely, interaction increases transparency in political affairs and the involvement of citizens in political decision-making processes (Stieglitz & Dang-Xuan, 2013). Moreover, interaction can entice new people in politics (Vergeer, Hermans & Sams, 2011). For example, through interaction on social media politicians are able to reach especially youngsters, and make them enthusiastic about politics and encourage them to vote (Moeller, De Vreese, Esser & Kunz, 2014).
Finally, discussing politics with others online is a way for youngsters in particular to increase political efficacy (Moeller et al., 2014).
Graham, Jackson and Broersma’s (2014) study on candidate’s use of Twitter during election campaigns showed that Twitter is becoming a place where interaction between politicians and citizens can evolve. They suggest that future research should delve into the use of Twitter by citizens with regard to their interaction with politicians. Graham, Broersma, Hazelhoff and Van ‘t Haar (2013) investigated with whom candidates interact on Twitter, but not which content actually encourages interaction. In addition, Druckman (2004, p. 15) suggested that interpersonal discussions are capable of shaping citizens’ voting decisions and that future research would benefit from exploring this subject. Adding to this existing knowledge of politicians’ use of Twitter and the discussion on the use of interactivity in political campaigning, this research focuses on what content in messages on Twitter sent by list pullers can trigger interaction from other Twitter users. Therefore, the research question that will be answered is the following:
Which content from list pullers’ tweets creates interaction on Twitter?
8
2. Theoretical framework
The following chapter gives an overview of the existing relevant literature on political communication and interaction on social media. At first, an overview of the history of political communication will be given, followed by a second section on the use of social media in a political context. The third section discusses what is already known about interaction on social media, the advantages and disadvantages of political communication on social media, and the aspects of interaction with regard to Twitter.
2.1 Political communication
In many democracies, political communication has gone through three phases during the post-war period. The first period was during the first two decades after World War II, and is called “the ‘golden age’ of parties” (Blumler & Kavanagh, 1999, p. 211). During that period, the political arena was dominated by strong parties that were able to get their messages to the media without much difficulty, and they were supported by loyal voters (Blumler & Kavanagh, 1999). It was with difficulty that citizens were able to select the sources that reflected their own political preferences, as the range of sources was very limited (Bennett & Iyengar, 2008). Citizens were only able to expose themselves to content considering preferred parties and candidates during campaigns (Bennett &
Iyengar, 2008). This type of campaigning was characterized by the use of newspapers and direct face- to-face communication during rallies and meetings (Vergeer, Hermans & Sams, 2011).
During the 60’s, the second period began with the arrival of the limited-channel nationwide television (Vergeer, Hermans & Sams, 2011). Politics were brought into the living rooms of citizens who thereby became more involved in politics (Gurevitch, Coleman & Blumler, 2009). Through television more people were reached than before, enticing new people in politics. Television became a dominant medium for political parties to broadcast their messages. Gurevitch, Coleman, and Blumler (2009) call this the television-politics relationship in which television journalists depend on the content provided by politicians, and politicians depend on being broadcasted. The voters’ loyalty to one party was loosening in this period, which was, among other things, due to the less selective news channels that provided voters with a broader scope on politics (Blumler & Kavanagh, 1999). This broader scope included, among other things, recent events and governments’ successes and failures.
The third phase is characterized by the broad availability of television and radio channels for political communication (Blumler & Kavanagh, 1999). The computer and the Internet were also introduced in the third phase, both allowing people to search for information and engage in discussions beyond the mass media twenty-four-seven (Blumler & Kavanagh, 1999). The broad availability of media channels led to a more competitive environment in which politicians have to compete for attention from journalists as well as audiences (Blumler & Kavanagh, 1999).
9 2.2 Political communication and social media
This broad online media environment has many users. For example, in 2011, the platform Facebook had over 800 million users worldwide, and Twitter had 200 million users (Stieglitz & Dang-Xuan, 2013).
In January 2017, there were over 1,870 million active Facebook users and over 317 million active Twitter users (Allen, 2017), which shows immense growth of both platforms. On these social media platforms all users are capable of publishing their own content, micro-blogs and weblogs (Stieglitz &
Dang-Xuan, 2013). In addition to the possibility to generate one’s own content online, the Internet allows users to cooperate, share content, socialize, and network with other users (Stieglitz & Dang- Xuan, 2013; Vergeer & Hermans, 2013). This results in an ever increasing variety and amount of content regarding political affairs on social networking sites (SNSs).
The academic discipline of political communication had already thoroughly researched the traditional mass media, before the rise of the Internet. These days, the Internet has established itself next to the mass media (Dahlgren, 2005). Compared to traditional mass media, the study of online political communication is interesting in that the Internet allows the presence of many more political voices, more ways for political engagement, and an increase in definitions of what constitutes politics (Dahlgren, 2005). However, research also has to take into account the information overload that results from the access of a seemingly limitless amount of sources provided by, among others, politicians, political parties, and individual bloggers (Bennett & Iyengar, 2008).
2.2.1 Use of social media by politicians
With that many spaces of mediation and the growth of social media platforms, there are consequences for politicians. Namely, these days, politicians are forced to engage in
“multidimensional impression management” in the broad media environment (Gurevitch, Coleman &
Blumler, 2009, p. 173). Thus, politicians started participating on these platforms as well, which led to the presence of many politicians and citizens on SNSs (Stieglitz & Dang-Xuan, 2013).
When politicians use social media, they do that independently. Because the politician is the sender, a politician no longer necessarily depends on assistance by, among others, party officials or journalists determining which events are deemed newsworthy (Klinger & Svensson, 2014; Vergeer, Hermans & Sams, 2011; Vergeer & Hermans, 2013). Furthermore, politicians are able to send as many messages as they would like at any given moment, as SNSs offer fast communication channels at very low costs that are unhindered by national or geographical boundaries (Stieglitz & Dang-Xuan, 2013;
Vergeer & Hermans, 2013).
Politicians also use social media because SNSs are networked (Vergeer, Hermans & Sams, 2011). News travels fast within well-connected networks, which is important when a politician wishes
10
to spread a message. The network of a politician serves as a community from which a politician can gain support (Stieglitz & Dang-Xuan, 2013). Within this community, politicians are able to deploy an individualized and personal campaign strategy (Vergeer & Hermans, 2013). Such a strategy could lead to decreasing the psychological distance between politicians and citizens, which again increases the sense of community within the network. Some popular SNSs for individualized campaigning among politicians are Facebook, YouTube, and Twitter (Vergeer, Hermans & Sams, 2011).
2.2.2 Use of social media by citizens in a political context
Social media are part of what Downey and Fenton (2003) call non-mass media or community media. A relevant characteristic of these media is that the production of content is often based on participation by citizens (Downey & Fenton, 2003). In a political context, this results in users of blogs discussing political affairs with other individuals, spreading their political opinion to a wider audience (Baum &
Groeling, 2008). With regard to this, it is important to note that users are more likely to share content from sources with a similar ideology instead of content from dissimilar sources (Barberá, Jost, Nagler, Tucker & Bonneau, 2015). Besides discussing and sharing political views, being pro-active online in finding political information, and engaging in campaigns leads to citizens feeling better informed, experiencing political efficacy, and being willing to participate in democratic processes (Gurevitch, Coleman & Blumler, 2009).
2.2.3 Disadvantages of social media in a political context
It is important to acknowledge that the immense growth of social media use and the platforms themselves have some important drawbacks with regard to political communication and political information gathering. The first drawback is that using the Internet for political information gathering generally leads to users avoiding opinion-challenging content and only selecting the information that confirms users’ existing points of view (Halberstam & Knight, 2016; Knobloch-Westerwick & Meng, 2009). Consequently, users are unable to form an informed opinion based on a variety of viewpoints, which in turn leads to a more polarized and divided electorate, and to a reduction of political tolerance (Knobloch-Westerwick & Meng, 2009). Klinger and Svensson (2014) refer to the avoidance of opinion-challenging content as selective exposure. Because of this, politicians tend to reach a self- selected audience instead of a general public, and are therefore not addressing new potential voters (Klinger & Svensson, 2014). The second drawback refers to information overload, as the Internet offers an unparalleled amount of easily accessible information to process (Gurevitch, Coleman &
Blumler, 2009). In continuation of this, users of the Internet are uncertain of which information they can trust (Gurevitch, Coleman & Blumler, 2009).
11 2.2.4 Advantages of social media in a political context
Besides drawbacks, using the Internet and social media for political information sharing and gathering comes with many advantages. First, the Internet is beneficial as a source of political information in that it operates twenty-four-seven (Johnson & Kaye, 2000). Second, social media offer users efficient communication at low costs (Kaplan & Haenlein, 2010). Third, sharing information on the Internet is free of the professional and social constraints to provide readers with an accurate and unbiased overview of events, as opposed to television and newspapers, allowing users to share their own opinions without difficulty (Johnson & Kaye, 2000). Fourth, SNSs provide (young) voters with platforms to discuss politics, share information and form an opinion, which increases their internal efficacy with regard to politics (Moeller et al., 2014). Fifth, social media platforms can be used to organize groups (Laroche, Habibi, Richard & Sankaranarayanan, 2012); creating tight communities of followers of a politician. Although the opportunities are not yet being exploited to the full extent, social media platforms offer a relevant sixth benefit. This opportunity regards a feature of SNSs that in the political context can provide more participation and democracy (Stieglitz & Dang-Xuan, 2013), and political engagement (Vergeer, Hermans & Sams, 2011). These are the interactive features of SNSs which allow politicians to directly interact with citizens on social media platforms (Stieglitz & Dang-Xuan, 2013;
Vergeer, Hermans & Sams, 2011).
2.3 Interactivity
Social media platforms offer a wide range of possibilities to stimulate interaction between users.
However, as mentioned above, research shows that the opportunity to interact on social media remains unexploited (Vergeer, Hermans & Sams, 2011). In addition, it appears that political websites are mostly used in the same manner as the traditional mass media, resulting in one-way communication, ignoring among other things the potential for interactivity and horizontal communication (Vergeer & Hermans, 2013). Also, the websites only reach users who actively search for them. Therefore, examples of good practice of online interaction by politicians are scarce, although many politicians claim that it is of great importance that governments listen and converse with citizens (Gurevitch, Coleman & Blumer, 2009). After all, a meaningful relationship between citizens and politicians is important, as citizens need to feel represented by politicians in order for a democratic government to be successful (Graham, Broersma, Hazelhoff & Van ‘t Haar, 2013).
Optimism regarding interaction on SNSs still exists, as Vergeer and Hermans (2013) show that the first signs of an increase in interactive behaviour were found in earlier studies. For example, Vaccari (2008) showed that in the 2004 US elections candidates used an email list with committed volunteers and supporters, which could be used to create a virtual community for virtual campaigning.
12
Another example comes from Foot, Schneider and Dougherty (2007) who show that the 2004 US congressional campaign of Howard Dean featured a network of websites that connected over 500 discussion groups, action coordinators, and supporters’ websites, creating a huge community in which users could interact with each other. These examples of good practice might indicate that the interactive features of SNSs will be deployed more in the future.
Engaging in interaction would take a shift in the manner in which politicians broadcast their messages. Whereas politicians used to have control over the political agenda and be proactive, they are now forced to be more responsive (Gurevitch, Coleman & Blumler, 2009). These days, politicians need to adapt to the interactive audience by responding to their questions and challenging messages, redistribute messages, and modify received messages, while also appearing as a sincere and authentic person with whom citizens would want to interact (Gurevitch, Coleman & Blumler, 2009).
2.3.1 Benefits of interaction
Participating in the conversations on SNSs entails advantages for both politicians and citizens. First, by interacting with citizens, politicians keep citizens up to date and they give citizens attention. These are things that citizens desire from politicians, and therefore politicians might earn more votes if they fulfil this desire (Spierings & Jacobs, 2014). Second, if politicians would interact with citizens on social media platforms, this might lead to more transparency in political affairs, and involvement of citizens in processes of political decision-making (Stieglitz & Dang-Xuan, 2013). Furthermore, by using Twitter more actively and by engaging in interaction on Twitter, politicians can reach an interesting cohort of citizens; the digital natives (Moeller et al., 2014). This is the youngest cohort of voters and these voters use SNSs in large quantities. Therefore, it is easier to reach them via social media, whereas they are far more difficult to reach via the more traditional mass media (Moeller et al., 2014; Vergeer, Hermans & Sams, 2011). Reaching these younger citizens through a medium they are already using, could entice them into the political realm (Vergeer, Hermans & Sams, 2011). Even more so, according to Moeller et al. (2014), younger citizens develop their internal political efficacy by engaging in political discussions and by sharing information about politics. As politicians can obtain votes from younger citizens and younger citizens experience political efficacy from interaction between politician and citizens, interaction is a double-edged sword.
2.4 Aspects of interaction on Twitter
In order to determine what constitutes interaction, and what kind of messages evoke interaction, it is interesting to consider the content and the structure of these messages in depth. In their article on social media analytics in political communication, Stieglitz and Dang-Xuan (2013) describe three
13
categories for content analysis of social media that can be applied individually or combined; the topic and issue category, the opinion and sentiment category, and the structural category. This section shows the aspects of interaction on the social medium Twitter using these three content categories.
2.4.1 Topic and issue category
The first content category from the work of Stieglitz and Dang-Xuan (2013) is the topic and issue category, which refers to the identification of the topic of a message. To get a clear view of the subject of a message, the content is determined by three variables; the topic of a tweet, possibly visual content, and tweet characteristics such as hashtags and @-mentions. Each of these aspects will be elaborated below.
Topic of a tweet
When studying political communication, it is possible to determine the presence of a political conversation by using a range of subjects. These subjects on the Internet and social media can be public as well as private (Shirky, 2011). Fernandes, Giurcanu, Bowers and Neely (2010) studied public content in their research. The subjects they used were based on Sweetser Trammell’s (2007) study, which were war, economy, security/defence, satisfaction and dissatisfaction with the government, international issues/foreign policy, education, and health care. These topics partially overlap with the overview of the Dutch party programmes (Kamerbreed, n.d.). Kamerbreed (n.d.) added the topics Europe, social affairs and employment, media and culture, integration, and citizen and governance.
Kamerbreed (n.d.) also contains a topic similar to health care, which is named public health, welfare and sports, and the topic economy includes taxes and other financial affairs.
There can also be private content on the Twitter accounts of politicians. Since Twitter provides users with the possibility to create their own content, and because politics have become more personalized, it is very likely that politicians also share content from their personal life on Twitter (Bennett, 2012). Sharing things from politicians’ private lives is beneficial, in that it engages citizens in the lives of the politicians. In their research concerning celebrities on Twitter, Marwick and Boyd (2011) call this performative intimacy. Private content relates to topics addressed by politicians considering their life outside of the political arena. Aspects of personal life to consider are family and friends, voluntary activities, religion, home, and leisure (e.g. sports and hobbies) (Chalofsky &
Cavallaro, 2013). Based on the argument by Marwick and Boyd (2011), it can be expected that Twitter users show a lot of interest in the more private tweets from politicians in the form of reactions, retweets, and likes.
14 Visual content
Users of Twitter are able to share ‘visual content’ alongside the 140-character messages. Pictures on Twitter are mostly related to everyday life, showing food, the weather, street scenes, and events (Kaneko & Yanai, 2013). In the context of political communication, such pictures would reveal things from the life of a politician. This could concern not only formal publicity stills and campaign material, but personal and candid pictures as well. Sharing personal pictures would even reinforce performative intimacy, offering citizens a glimpse into the personal life of the politician (Marwick & Boyd, 2012).
Visual content offers Twitter users a richer view of the life of celebrities, such as politicians, and because users are interested in those lives (Marwick & Boyd, 2012), it can be expected that the addition of visual content to a tweet will lead to more interaction.
Tweet characteristics
On the social media platform Twitter, a user has several features available to engage in interaction and create connections. Zappavigna (2011) describes these features, calling them ‘linguistic markers’.
Zappavigna (2011, p. 790) writes that these features can be used “to bring other voices into tweets by addressing other users, republishing other tweets, and flagging topics that may be adopted by multiple users”. The first refers to the @-mention, indicating that someone is addressed in the message by putting the username of the addressee behind the ‘@’ symbol (Zappavigna, 2011). The second type of interaction can be achieved by redistributing a message of another user with a retweet.
A tweet from a user will be shown on the feed of the user that retweets the message. A retweet can be recognized by the letters ‘RT’ in front of the tweet, and is often followed by a @-mention to indicate the source (Zappavigna, 2011). Finally, flagging topics can be done by using hashtags, which can be recognized by the ‘#’ symbol (Zappavigna, 2011). With a hashtag, the user defines the topic of the tweet and creates a reference to other tweets with the same hashtag (Zappavigna, 2011). Finally, users can also add URL’s, emoticons, and polls to their tweets. Polls invite other users to answer a multiple choice question. Based on this section, it can be expected that tweets with these features will bring about more interaction as compared to tweets that do not have these features.
2.4.2 Opinion and sentiment category
The second content category from Stieglitz and Dang-Xuan (2013) refers to the opinion and sentiment of a message. Users of social media can express, among other things, their points of view and feelings on social media. Users do so more than ever before, which is important, because people prefer to hear other opinions before they make their own decision (Stieglitz & Dang-Xuan, 2013). Within the opinion and sentiment category there are two important aspects to consider; the ‘tone’ of a message and the use of ‘humour’.
15 Tone
In research, opinion sharing is translated into ‘sentiment analysis’ or ‘opinion mining’ (Stieglitz & Dang- Xuan, 2013). This method is named coding for ‘tone’, or understanding “the valence of sentiment” in Diakopoulos and Shamma’s (2010, p. 1196) research. The codes for ‘tone’, that Diakopoulos and Shamma (2010) used in their study were negative, positive, mixed (positive as well as negative), or other (non-evaluative content). Fernandes et al. (2010) considered tone in terms of positive, negative, an equal mix of positive and negative, or neutral. In their analysis of a sample of political news articles, De Vreese et al. (2006) coded ‘tone of the news’ using the codes neutral (non-evaluative content), negative, positive, dominantly negative, dominantly positive, or mixed. Although computerized methods for analysing ‘tone’ have been greatly advanced, these methods still lack the ability to handle emoticons, acronyms, amplifications, slang, and sarcasm or irony in informal messages (Stieglitz &
Dang-Xuan, 2013).
Humour
In addition to ‘tone’, it is important to take ‘humour’ into consideration when analysing the sentiment of tweets (Raz, 2012; Zhang & Liu, 2014). This is important, because besides affecting feelings, humour also has an influence on human beliefs (Raz, 2012). The aspect of influencing human beliefs is important in the political context, as political messages from politicians as well as citizens aim at convincing others of, for example, the verity of a particular viewpoint. On Twitter, humorous posts possess certain characteristics that plain tweets and humorous non-tweets do not (Raz, 2012; Zhang &
Liu, 2014). Raz (2012) describes three theories of humour in order to recognize ‘humour’ in a tweet, the first being incongruity humour which refers to the presence of one statement with two contradictory interpretations as a condition for humour. The second is the superiority theory which involves feelings of victory or triumph over someone who is or something that is wrong, inferior, or defeated (Meyer, 2000; Raz, 2012). The third is the relief humour which refers to humour containing taboo and is described as “a license for banned thoughts” (Raz, 2012, p. 78). The humour releases physiological tension (Meyer, 2000). Based on the former paragraph, ‘humour’ in a tweet most likely leads to more interaction as it is convincing and appealing when used appropriately.
2.4.3 Structural category
The third and final content category for social media analytics is the structural category (Stieglitz &
Dang-Xuan, 2013). This category regards the identification of influential users (i.e. opinion leaders) of social media (Stieglitz & Dang-Xuan, 2013). Politicians might (attempt to) interact with such actors, or mention them in their messages (Stieglitz & Dang-Xuan, 2013). Thus, the relevant aspect of the structural category is ‘actors’.
16 Actors
As Zappavigna (2011) explained, Twitter users can use a @-mention to involve other users in a tweet.
This type of Twitter behaviour can take place between many different actors (Dahlgren, 2005). For example, mentioning other users can take place between citizens, but also, between citizens and the media, or politicians (Dahlgren, 2005). In their analyses of news coverage during the 2004 European parliamentary elections, De Vreese, Banducci, Semetko and Boomgaarden (2006) identified stories about the elections based on a set of codes defining different ‘types of actors’. In their research, an actor is a person, groups of persons with a shared interest, an institution, or another organization (De Vreese et al., 2006). It is interesting to investigate the ‘actors’ mentioned in a tweet, as some actors could generate more interaction than others. It might be expected that mentioning more influential Twitter users, such as other politicians, the media, or other opinion leaders, brings about more interaction than less influential users, such as citizens.
2.4.4 Additional content
Besides the three content categories that Stieglitz and Dang-Xuan (2013) describe, they also describe other aspects of a message that are interesting to take into account when executing an analysis of social media. This concerns an identification of the author of the message, and a time stamp. The following section shows two relevant aspects of a message; ‘network characteristics’ and ‘candidate characteristics’, and ‘timing’.
Network and candidate characteristics
In their research on the use of Twitter by candidates of the Dutch general elections, Vergeer and Hermans (2013) took into account ‘network characteristics’ and ‘candidate characteristics’. Doing so, contributes to the outlining of the interaction on Twitter based on individual candidates. Vergeer and Hermans (2013) measured ‘network characteristics’ by the network size, represented by the amount of followers of a politician, the amount of people the politician follows, and reciprocal following. A politician with more followers will most likely trigger more interaction with a tweet than a politician with fewer followers.
‘Candidate characteristics’ were measured by Vergeer and Hermans (2013) based on the prioritization of each candidate, meaning a politicians’ position on the list of electoral candidates of a party. The prioritization shows the likelihood that a candidate will be elected, with a lower number representing a higher prioritization. It is expected that, without considering the amount of followers, electoral candidates with a similar prioritization will bring about comparable amounts of interaction with their tweets.
17 Timing
Before the Internet, politicians and political parties decided, in cooperation with the media, when a message from that politician or political party would be distributed (Mangold & Faulds, 2009). Besides face-to-face and word-of-mouth communication, the receivers of the communication expressions were not in control of when a certain topic was distributed (Mangold & Faulds, 2009). However, with the possibilities of the Internet, everyone can access, (re)distribute, and comment on content at any given moment (Mangold & Faulds, 2009). Therefore, it is interesting for a politician to know on which moment it is most likely that people will see, redistribute, or comment on messages from that politician (De Vries, Gensler & Leeflang, 2012). When a politician has that knowledge, he or she is in the position to post whenever he or she can expect the most reactions, retweets, and likes, thus increasing popularity (De Vries, Gensler & Leeflang, 2012). In other words, if a politician has good
‘timing’, the politician gains back some control over which topics are discussed at what moments.
Based on the former paragraph it is expected that although users can access any public information on social media at any time, there will be moments during which a message will receive more response than messages sent at a different time.
2.5 Research goal
The former chapter offered insight in, among other things, the use of social media in political communication, the benefits and drawbacks of social media in political communication, the benefits of engaging in interaction, and which aspects of a message to consider when investigating interaction.
In this study, the use of these aspects of interaction by Dutch list pullers will be investigated in order to determine which Twitter content triggers interaction in a political context.
18
3. Method
An already existing sample of Twitter messages from politicians was analysed by means of a content analysis. Content analysis is considered a straightforward method to go through substantial amounts of data, and is very helpful in attempts to find patterns and trends (Stemler, 2001). The analysed corpus was gathered using NodeXL.
3.1 Context
The Netherlands started as a frontrunner in the adoption of social medium Twitter with an adoption rate of 22.0% of the Dutch population in 2010 (Vollman, 2011), and 27.0% in 2011 (Graham, Jackson &
Broersma, 2016; Vergeer & Hermans, 2013). The presence of such a large amount of Dutch citizens on Twitter led to the deployment of this micro-blogging service by many Dutch politicians. Today, in 2017, approximately 15.3% of the Dutch citizens have a Twitter account (Van der Veer, Boekee &
Peters, 2017), which indicates a decrease in the adoption rate. However, all list pullers of the former Dutch House of Representatives (i.e. Second Chamber) as well as the list pullers of new parties present in the current House of Representatives still use Twitter as an important part of their campaigns.
3.2 Corpus
The corpus of the research contained the original tweets sent by the list pullers who used Twitter in their campaign during the most recent Dutch election period (N = 2158), which ran between December 19th, 2016 and March 23rd, 2017 (Tweede Kamer der Staten-Generaal, n.d.). An original tweet is the first tweet of a conversation, and therefore not a reaction to another tweet or a retweet.
The elections determined which politicians and political parties would represent the Dutch people for the following four years (Kieswet 2001, art. C 1.1). An election period was chosen, since such a period is one of the most intensive with regard to communication and interaction between politicians and citizens (Graham, Jackson & Broersma, 2014). The selected list pullers for the research were all list pullers present in the former and the current House of Representatives (N = 13). Table 1 shows an overview of these list pullers, the party they are associated with, the number of tweets they sent during the election period, the amount of followers of each list puller, and the amount of users each list puller follows.
19
Table 1: Overview of selected list pullers participating in the 2017 Dutch elections
List puller Twitter username
Political
party Political orientation1
Original tweets during election
period Followers3
Following users3
Sybrand Buma @sybrandbuma CDA Centre right, conservative 31 70106 326
Tunahan Kuzu @tunahankuzu DENK Left, progressive 44 29152 68
Emile Roemer @emileroemer SP Left, progressive 54 175393 725
Mark Rutte @MinPres VVD Right, conservative 56 797875 0
Jesse Klaver @jesseklaver GL Centre left, progressive 58 89327 524
Alexander Pechtold @APechtold D66 Centre, progressive 90 630168 595
Kees van der Staaij @keesvdstaaij SGP Centre right, conservative 96 51975 1787
Marianne Thieme @mariannethieme PvdD Left, progressive 145 71746 3454
Gert-Jan Segers @gertjansegers CU Centre 209 22742 495
Lodewijk Asscher @LodewijkA PvdA Centre left, progressive 212 225744 1398
Henk Krol @HenkKrol 50PLUS Left 294 15488 218
Thierry Baudet @thierrybaudet FvD - 407 40556 193
Geert Wilders @geertwilderspvv PVV Centre right, conservative 4622 829760 1
1 From Kieskompas (2017)
2 For Geert Wilders, all English tweets that were direct translations of a Dutch tweet were deleted from the corpus
3 Numbers were retrieved on June 21st, 2017
3.3 Codebook
The codebook was constituted following the deductive approach, meaning that it was determined before the actual coding began (Semetko & Valkenburg, 2000; White & Marsh, 2006). This is beneficial, as the deductive approach is easy to replicate and is applicable to large samples (Semetko
& Valkenburg, 2000). The first codes of the codebook (Appendix A) are typical data to code, namely the ‘ID of a post’, the ‘timing’ of the post, and a reference to the author in the form of the ‘network characteristics’ and ‘candidate characteristics’ (Stieglitz & Dang-Xuan, 2013). The codebook is based on the three content categories from the work of Stieglitz and Dang-Xuan (2013) described in the theoretical framework. In their research, they describe a guideline for developing toolsets and codebooks for the analysis of social media in a political context (Stieglitz & Dang-Xuan, 2013). The following two sections show the interaction variables of the research based on the features of Twitter, and the independent variables falling under the three content categories.
20 3.3.1 Interaction variables
The interaction variables apply particularly to the features of Twitter. The dependent variables of this research were the interaction variables shown in Table 2; amount of reactions (number of reactions following the original tweet), amount of retweets (number of times the original tweet was redistributed), and amount of likes (number of likes for the original tweet).
Table 2: Frequencies of interaction variables in corpus
Dependent variable Minimum Maximum Mean Median Std. deviation
Amount of reactions 0 1471 47.30 13.50 103.807
Amount of retweets 0 5823 142.95 34.00 340.329
Amount of likes 0 9536 246.97 52.00 554.945
3.3.2 Independent variables
The following elaborates upon the independent variables of the codebook. Each of those variables is a code in the codebook which falls under the three content categories from Stieglitz and Dang-Xuan (2013); the topic and issue category, the opinion and sentiment category, and the structural category.
Topic and issue category
The topic and issue category refers to the content of an original tweet sent by a list puller. First, the
‘topic of a tweet’ was coded by choosing the most relevant and prominent topic from a list of topics based on literature (e.g. war/terrorism, education, health care). Additional topics were campaign activities, because the tweets were sent during the election period, and celebration for national holidays that took place during that same period, and other. Finally, for tweets that addressed more than one topic without one of them standing out the most, the code multiple topics was used. Second, when a tweet contained ‘visual content’ (e.g. formal publicity, street scenes, events) it was also coded for the most relevant and prominent subject. If a tweet did not contain any visual content, it was coded not applicable. Third, besides a topic and visual content, other ‘tweet characteristics’ could be hashtags (#), @-mentions, polls, emoticons, and URL’s, which were coded using no (0) or yes (1).
Opinion and sentiment category
To represent the opinion and sentiment category, this research measured opinion and sentiment by coding for ‘tone’ of a tweet (negative, non-evaluative, positive, and mixed) and the presence of
‘humour’ (no or yes) in a tweet. ‘Tone’ indicated the emotional valence of a tweet. A tweet was coded as negative if it showed emotions such as sadness, anger, and confusion, whereas positive tweets
21
showed happiness, satisfaction, excitement, or curiosity. Non-evaluative tweets were neutral and showed none of these emotions. A tweet was coded as mixed if both positive as well as negative emotion was present in a tweet. Also, the use of emoticons could reinforce the emotional valence of a tweet, which was useful for indicating the ‘tone’. ‘Humour’ was coded as present when a tweet contained, for example, jokes, wordplay, sarcasm, or irony. A winking emoticon could indicate use of
‘humour’ as well, and therefore extra attention was paid to tweets containing a winking emoticon.
Structural category
The structural category was applied by coding the ‘actors’ (e.g. a citizen, a politician, media) mentioned by the politicians in their original tweet by using a @-mention. In order to find out which type of actor a mentioned user was, coders first looked at the user’s profile, and if it was necessary to the URL in the user’s account description.
3.4 Validity and reliability
In order to ensure the codebook’s validity, it contained categories that were relevant in answering the research question and that only measured the intended concept (Stemler, 2001; White & Marsh, 2006). Thus, categories had to be mutually exclusive, meaning that data could not fall between two categories and all data was represented by only one category, and exhaustive, meaning that all important aspects of a category are represented in the data (Stemler, 2001; White & Marsh, 2006).
To ensure the reliability and reproducibility of this study, it was important that all coders would code the same item in the same manner. Because the research applied the deductive approach, it was possible to pre-test the codebook to control for the coding behaviour of the researcher, and thus ensuring the reliability of the research. The pre-test was executed by appointing a second coder to code a random selection of 10.2% (N = 221) of the corpus using the codebook. The researcher coded the same selection. After coding the tweets, the codes of the second coder were compared to the codes of the researcher, or first coder, using Cohen’s Kappa. With an average inter-coder agreement of .7, it was sufficiently reliable. Only one pre-test was executed. The Kappa’s from the pre- test are shown in Table 3, and the separate Kappa’s for each actor are shown in Appendix B.
22 Table 3: Results pre-test
Based on the first and only pre-test of this research, not all individual codes were reliable because the Kappa’s were not high enough. This was true for three of the fifteen codes for ‘actors’; politician sender’s party (k = 0.534, 95% CI, p < .005), interest group (k = -0.006, 95% CI, p = .924), and other actors (k = 0.349, 95%, p < .005), and to the codes ‘tone’ and ‘humour’.
Codes with an insufficient Kappa were adapted or the descriptions in the codebook were improved in accordance with consultation with the second coder. This resulted in more examples of actors in the description of the codes for ‘actors’, and to more examples of what types of humour a humorous tweet could contain. The code ‘tone’ was adapted to be more unambiguous by removing the vaguer codes slightly negative and slightly positive. Finally, based on the pre-test it became evident that the codes for ‘visual content’ and codes for ‘topic of the tweet’ were not exhaustive.
Therefore text and media was added to the code ‘visual content’, and the categories environment, animals, and public transportation/infrastructure were added to the code ‘topic of the tweet’.
3.5 Data analysis
The unit of coding consists of one individual tweet. This refers to each original tweet sent by a list puller from the former and current House of Representatives during the Dutch elections of 2017.
Coding and analysis were performed using the statistics programme SPSS. In order to determine which content from the list pullers was followed by a significantly large amount of interaction, a median split of the interaction variables was performed. Using a median split facilitates the interpretation of the
Code
Initial
Kappa Sig.
Confidence interval
Interaction variables Reactions 0.866 p < .005 95%
Retweets 0.917 p < .005 95%
Likes 0.881 p < .005 95%
Topic and issue category Topic of the tweet 0.503 p < .005 95%
Visual content 0.738 p < .005 95%
Emoticon 0.829 p < .005 95%
Hashtag 0.929 p < .005 95%
@-mention 0.932 p < .005 95%
Poll1 - - -
URL 0.753 p < .005 95%
Opinion and sentiment category Tone 0.373 p < .005 95%
Humour -0.025 p = .705 95%
Structural category Actors (mean) 0.687 p < .005 95%
1 Not present in pre-test
23
results. This split divided the interaction variables, creating a part with low interaction and a part with high interaction. This resulted in an approximately equal distribution of the interaction variables
‘reactions’ (low = 1115, high = 1043), ‘retweets’ (low = 1086, high = 1072), and ‘likes’ (low = 1091, high
= 1067) as shown in Table 4.
Table 4: Median split of interaction variables
Reactions Retweets Likes
Frequency Percentage Frequency Percentage Frequency Percentage
Low 1115 51.7 1086 50.3 1091 50.6
High 1043 48.3 1072 49.7 1067 49.4
Total 2158 100.0 2158 100.0 2158 100.0
The split variables were used in cross tables and Chi²-tests to determine if an independent variable triggered significantly less or more ‘reactions’, ‘retweets’, and ‘likes’. The residuals in the cross tables showed a significant effect if those numbers were higher of lower than 2 (Lammers, Pelzer, Hendrickx
& Eisinga, 2007).
24
4. Results
In total, 2158 original tweets from 13 list pullers seated in the former and/or current Dutch House of Representatives were analysed using the codebook (Appendix A). It was first investigated if the interaction variables ‘reactions’, ‘retweets’, and ‘likes’ correlated with each other. If they do, it means that when one interaction variable is high or low, it is very likely that the other interaction variables are high or low as well. As can be seen in Table 5, the interaction variables correlate with each other on a moderate to high level.
Table 5: Correlation between interaction variables
Reactions Retweets
Reactions Pearson Correlation 1 -
Sig. (2-tailed) - -
Retweets Pearson Correlation 0.736 1
Sig. (2-tailed) < .001 -
Likes Pearson Correlation 0.768 0.895
Sig. (2-tailed) < .001 < .001
The results will be presented in the sections below, following the three content categories of Stieglitz and Dang-Xuan (2013). The chapter closes with the results of the codes ‘network characteristics’ and
‘candidate characteristics’, and ‘timing’.
4.1 Topic and issue category
The topic and issue category is represented by the independent variables ‘topic of a tweet’, the ‘visual content’ that is possibly added to a tweet, and the ‘tweet characteristics’ in the form of emoticons, hashtags, @-mentions, polls, and URL’s.
4.1.1 Topic of the tweet
With regard to the ‘topic of the tweet’, Figure 1 shows that during the election period, list pullers mostly sent out tweets with regard to their campaign activities (N = 702).
25 Figure 1: Frequencies of topic of the tweet
Besides campaign activities, other popular topics were media and culture (N = 170), commenting on the government and other politicians (N = 170), and social affairs and employment (N = 113). Public subjects that were mentioned the least often were environment (N = 24), public transportation/infrastructure (N = 18), and citizen and governance (N = 10). Tweets considered mostly public content. For example, topics such as leisure/sports/hobbies (N = 39), home (N = 9), and family/friends (N = 9) have relatively low frequencies as compared to most public topics.
A Chi²-test was executed to find significant differences in the amount of interaction triggered by the topic of a tweet. These differences were found for ‘reactions’ (χ²(22, N = 2158) = 202.008, p <
.001), for ‘retweets’ (χ²(22, N = 2158) = 221.705, p < .001), and for ‘likes’ (χ²(22, N = 2158) = 245.940, p
< .001). The residuals showed that the topics that triggered significantly more reactions, retweets and likes were war/terrorism, comments on government, international issues/foreign policy, Europe, integration/refugee policy, religion, and tweets with multiple topics. Topics that resulted in significantly fewer reactions, retweets and likes were economy/financial affairs/taxes, health care, social affairs and employment, media and culture, and public transportation/infrastructure. The topic campaign activities led to significantly fewer reactions and retweets (Table 6).
153 51
9 9 10
18 24 26 34
39 40 43 46 54
66 82
87 104
108 113
170 170
702
0 100 200 300 400 500 600 700 800
Other Multiple topics Family/Friends Home Citizen and governance Public transportation/Infrastructure Environment Education Animals Leisure/Sports/Hobbies Religion War/Terrorism Security/Defence Integration/Refugee policy Celebration Europe Health care International issues/Foreign policy Economy/Financial affairs/Taxes Social affairs and employment Commenting on government Media and culture Campaign activities
